2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
42 #include "opt_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/mutex.h>
54 #include <sys/resourcevar.h>
55 #include <sys/syscall.h>
56 #include <sys/vnode.h>
59 #include <sys/ktrace.h>
60 #include <sys/kthread.h>
61 #include <sys/unistd.h>
67 #include <vm/vm_map.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_zone.h>
71 #include <sys/vmmeter.h>
74 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
77 * These are the stuctures used to create a callout list for things to do
78 * when forking a process
82 TAILQ_ENTRY(forklist) next;
85 static struct sx fork_list_lock;
87 TAILQ_HEAD(forklist_head, forklist);
88 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
90 #ifndef _SYS_SYSPROTO_H_
97 init_fork_list(void *data __unused)
100 sx_init(&fork_list_lock, "fork list");
102 SYSINIT(fork_list, SI_SUB_INTRINSIC, SI_ORDER_ANY, init_fork_list, NULL);
111 struct fork_args *uap;
117 error = fork1(td, RFFDG | RFPROC, &p2);
119 td->td_retval[0] = p2->p_pid;
120 td->td_retval[1] = 0;
133 struct vfork_args *uap;
139 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
141 td->td_retval[0] = p2->p_pid;
142 td->td_retval[1] = 0;
154 struct rfork_args *uap;
159 /* Don't allow kernel only flags. */
160 if ((uap->flags & RFKERNELONLY) != 0)
163 error = fork1(td, uap->flags, &p2);
165 td->td_retval[0] = p2 ? p2->p_pid : 0;
166 td->td_retval[1] = 0;
173 int nprocs = 1; /* process 0 */
175 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
179 * Random component to lastpid generation. We mix in a random factor to make
180 * it a little harder to predict. We sanity check the modulus value to avoid
181 * doing it in critical paths. Don't let it be too small or we pointlessly
182 * waste randomness entropy, and don't let it be impossibly large. Using a
183 * modulus that is too big causes a LOT more process table scans and slows
184 * down fork processing as the pidchecked caching is defeated.
186 static int randompid = 0;
189 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
194 error = sysctl_handle_int(oidp, &pid, 0, req);
195 if (error || !req->newptr)
197 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
199 else if (pid < 2) /* NOP */
201 else if (pid < 100) /* Make it reasonable */
207 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
208 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
212 kse_init(struct kse *kse1, struct kse *kse2)
217 thread_init(struct thread *thread1, struct thread *thread2)
222 ksegrp_init(struct ksegrp *ksegrp1, struct ksegrp *ksegrp2)
228 fork1(td, flags, procp)
229 struct thread *td; /* parent proc */
231 struct proc **procp; /* child proc */
233 struct proc *p2, *pptr;
235 struct proc *newproc;
238 static int pidchecked = 0;
241 struct proc *p1 = td->td_proc;
248 /* Can't copy and clear */
249 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
253 * Here we don't create a new process, but we divorce
254 * certain parts of a process from itself.
256 if ((flags & RFPROC) == 0) {
257 vm_forkproc(td, NULL, NULL, flags);
260 * Close all file descriptors.
262 if (flags & RFCFDG) {
263 struct filedesc *fdtmp;
264 fdtmp = fdinit(td); /* XXXKSE */
266 fdfree(td); /* XXXKSE */
272 * Unshare file descriptors (from parent.)
275 FILEDESC_LOCK(p1->p_fd);
276 if (p1->p_fd->fd_refcnt > 1) {
277 struct filedesc *newfd;
280 FILEDESC_UNLOCK(p1->p_fd);
286 FILEDESC_UNLOCK(p1->p_fd);
293 * Although process entries are dynamically created, we still keep
294 * a global limit on the maximum number we will create. Don't allow
295 * a nonprivileged user to use the last process; don't let root
296 * exceed the limit. The variable nprocs is the current number of
297 * processes, maxproc is the limit.
299 uid = p1->p_ucred->cr_ruid;
300 if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
305 * Increment the nprocs resource before blocking can occur. There
306 * are hard-limits as to the number of processes that can run.
311 * Increment the count of procs running with this uid. Don't allow
312 * a nonprivileged user to exceed their current limit.
314 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
315 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
318 * Back out the process count
324 /* Allocate new proc. */
325 newproc = zalloc(proc_zone);
328 * Setup linkage for kernel based threading
330 if((flags & RFTHREAD) != 0) {
331 newproc->p_peers = p1->p_peers;
332 p1->p_peers = newproc;
333 newproc->p_leader = p1->p_leader;
335 newproc->p_peers = NULL;
336 newproc->p_leader = newproc;
339 newproc->p_vmspace = NULL;
342 * Find an unused process ID. We remember a range of unused IDs
343 * ready to use (from lastpid+1 through pidchecked-1).
345 * If RFHIGHPID is set (used during system boot), do not allocate
348 sx_xlock(&allproc_lock);
349 trypid = lastpid + 1;
350 if (flags & RFHIGHPID) {
356 trypid += arc4random() % randompid;
360 * If the process ID prototype has wrapped around,
361 * restart somewhat above 0, as the low-numbered procs
362 * tend to include daemons that don't exit.
364 if (trypid >= PID_MAX) {
365 trypid = trypid % PID_MAX;
370 if (trypid >= pidchecked) {
373 pidchecked = PID_MAX;
375 * Scan the active and zombie procs to check whether this pid
376 * is in use. Remember the lowest pid that's greater
377 * than trypid, so we can avoid checking for a while.
379 p2 = LIST_FIRST(&allproc);
381 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
382 while (p2->p_pid == trypid ||
383 p2->p_pgrp->pg_id == trypid ||
384 p2->p_session->s_sid == trypid) {
386 if (trypid >= pidchecked)
389 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
390 pidchecked = p2->p_pid;
391 if (p2->p_pgrp->pg_id > trypid &&
392 pidchecked > p2->p_pgrp->pg_id)
393 pidchecked = p2->p_pgrp->pg_id;
394 if (p2->p_session->s_sid > trypid &&
395 pidchecked > p2->p_session->s_sid)
396 pidchecked = p2->p_session->s_sid;
400 p2 = LIST_FIRST(&zombproc);
406 * RFHIGHPID does not mess with the lastpid counter during boot.
408 if (flags & RFHIGHPID)
414 p2->p_stat = SIDL; /* protect against others */
416 LIST_INSERT_HEAD(&allproc, p2, p_list);
417 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
418 sx_xunlock(&allproc_lock);
421 * Make a proc table entry for the new process.
422 * Start by zeroing the section of proc that is zero-initialized,
423 * then copy the section that is copied directly from the parent.
425 td2 = thread_get(p2);
429 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start))
431 bzero(&p2->p_startzero,
432 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero));
433 bzero(&ke2->ke_startzero,
434 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero));
435 bzero(&td2->td_startzero,
436 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero));
437 bzero(&kg2->kg_startzero,
438 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero));
441 bcopy(&p1->p_startcopy, &p2->p_startcopy,
442 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy));
443 bcopy(&td->td_kse->ke_startcopy, &ke2->ke_startcopy,
444 (unsigned) RANGEOF(struct kse, ke_startcopy, ke_endcopy));
445 bcopy(&td->td_startcopy, &td2->td_startcopy,
446 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy));
447 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
448 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy));
453 * XXXKSE Theoretically only the running thread would get copied
454 * Others in the kernel would be 'aborted' in the child.
455 * i.e return E*something*
457 proc_linkup(p2, kg2, ke2, td2);
459 mtx_init(&p2->p_mtx, "process lock", MTX_DEF);
461 /* note.. XXXKSE no pcb or u-area yet */
464 * Duplicate sub-structures as needed.
465 * Increase reference counts on shared objects.
466 * The p_stats and p_sigacts substructs are set in vm_forkproc.
469 mtx_lock_spin(&sched_lock);
470 p2->p_sflag = PS_INMEM;
471 if (p1->p_sflag & PS_PROFIL)
473 mtx_unlock_spin(&sched_lock);
475 p2->p_ucred = crhold(p1->p_ucred);
476 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */
479 p2->p_args->ar_ref++;
481 if (flags & RFSIGSHARE) {
482 p2->p_procsig = p1->p_procsig;
483 p2->p_procsig->ps_refcnt++;
484 if (p1->p_sigacts == &p1->p_uarea->u_sigacts) {
485 struct sigacts *newsigacts;
489 /* Create the shared sigacts structure */
490 MALLOC(newsigacts, struct sigacts *,
491 sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
495 * Set p_sigacts to the new shared structure.
496 * Note that this is updating p1->p_sigacts at the
497 * same time, since p_sigacts is just a pointer to
498 * the shared p_procsig->ps_sigacts.
500 p2->p_sigacts = newsigacts;
501 *p2->p_sigacts = p1->p_uarea->u_sigacts;
506 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
507 M_SUBPROC, M_WAITOK);
510 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
511 p2->p_procsig->ps_refcnt = 1;
512 p2->p_sigacts = NULL; /* finished in vm_forkproc() */
514 if (flags & RFLINUXTHPN)
515 p2->p_sigparent = SIGUSR1;
517 p2->p_sigparent = SIGCHLD;
519 /* bump references to the text vnode (for procfs) */
520 p2->p_textvp = p1->p_textvp;
528 else if (flags & RFFDG) {
529 FILEDESC_LOCK(p1->p_fd);
531 FILEDESC_UNLOCK(p1->p_fd);
538 * If p_limit is still copy-on-write, bump refcnt,
539 * otherwise get a copy that won't be modified.
540 * (If PL_SHAREMOD is clear, the structure is shared
544 if (p1->p_limit->p_lflags & PL_SHAREMOD)
545 p2->p_limit = limcopy(p1->p_limit);
547 p2->p_limit = p1->p_limit;
548 p2->p_limit->p_refcnt++;
552 * Preserve some more flags in subprocess. PS_PROFIL has already
555 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
556 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
557 p2->p_flag |= P_CONTROLT;
558 if (flags & RFPPWAIT)
559 p2->p_flag |= P_PPWAIT;
561 LIST_INSERT_AFTER(p1, p2, p_pglist);
566 * Attach the new process to its parent.
568 * If RFNOWAIT is set, the newly created process becomes a child
569 * of init. This effectively disassociates the child from the
572 if (flags & RFNOWAIT)
576 sx_xlock(&proctree_lock);
580 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
581 sx_xunlock(&proctree_lock);
583 LIST_INIT(&p2->p_children);
584 LIST_INIT(&td2->td_contested); /* XXXKSE only 1 thread? */
586 callout_init(&p2->p_itcallout, 0);
587 callout_init(&td2->td_slpcallout, 1); /* XXXKSE */
592 * Copy traceflag and tracefile if enabled. If not inherited,
593 * these were zeroed above but we still could have a trace race
594 * so make sure p2's p_tracep is NULL.
596 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
597 p2->p_traceflag = p1->p_traceflag;
598 if ((p2->p_tracep = p1->p_tracep) != NULL) {
609 * set priority of child to be that of parent
610 * XXXKSE hey! copying the estcpu seems dodgy.. should split it..
612 mtx_lock_spin(&sched_lock);
613 p2->p_ksegrp.kg_estcpu = p1->p_ksegrp.kg_estcpu;
614 mtx_unlock_spin(&sched_lock);
617 * This begins the section where we must prevent the parent
618 * from being swapped.
625 * Finish creating the child process. It will return via a different
626 * execution path later. (ie: directly into user mode)
628 vm_forkproc(td, p2, td2, flags);
630 if (flags == (RFFDG | RFPROC)) {
632 cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
633 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
635 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
636 } else if (p1 == &proc0) {
638 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
641 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
645 * Both processes are set up, now check if any loadable modules want
646 * to adjust anything.
647 * What if they have an error? XXX
649 sx_slock(&fork_list_lock);
650 TAILQ_FOREACH(ep, &fork_list, next) {
651 (*ep->function)(p1, p2, flags);
653 sx_sunlock(&fork_list_lock);
656 * If RFSTOPPED not requested, make child runnable and add to
659 microtime(&(p2->p_stats->p_start));
660 p2->p_acflag = AFORK;
661 if ((flags & RFSTOPPED) == 0) {
662 mtx_lock_spin(&sched_lock);
665 mtx_unlock_spin(&sched_lock);
669 * Now can be swapped.
675 * tell any interested parties about the new process
677 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
681 * Preserve synchronization semantics of vfork. If waiting for
682 * child to exec or exit, set P_PPWAIT on child, and sleep on our
683 * proc (in case of exit).
686 while (p2->p_flag & P_PPWAIT)
687 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
691 * Return child proc pointer to parent.
698 * The next two functionms are general routines to handle adding/deleting
699 * items on the fork callout list.
702 * Take the arguments given and put them onto the fork callout list,
703 * However first make sure that it's not already there.
704 * Returns 0 on success or a standard error number.
709 forklist_fn function;
714 /* let the programmer know if he's been stupid */
715 if (rm_at_fork(function))
716 printf("WARNING: fork callout entry (%p) already present\n",
719 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
722 ep->function = function;
723 sx_xlock(&fork_list_lock);
724 TAILQ_INSERT_TAIL(&fork_list, ep, next);
725 sx_xunlock(&fork_list_lock);
730 * Scan the exit callout list for the given item and remove it..
731 * Returns the number of items removed (0 or 1)
736 forklist_fn function;
740 sx_xlock(&fork_list_lock);
741 TAILQ_FOREACH(ep, &fork_list, next) {
742 if (ep->function == function) {
743 TAILQ_REMOVE(&fork_list, ep, next);
744 sx_xunlock(&fork_list_lock);
749 sx_xunlock(&fork_list_lock);
754 * Handle the return of a child process from fork1(). This function
755 * is called from the MD fork_trampoline() entry point.
758 fork_exit(callout, arg, frame)
759 void (*callout)(void *, struct trapframe *);
761 struct trapframe *frame;
763 struct thread *td = curthread;
764 struct proc *p = td->td_proc;
766 td->td_kse->ke_oncpu = PCPU_GET(cpuid);
768 * Setup the sched_lock state so that we can release it.
770 sched_lock.mtx_lock = (uintptr_t)td;
771 sched_lock.mtx_recurse = 0;
773 td->td_savecrit = CRITICAL_FORK;
774 CTR3(KTR_PROC, "fork_exit: new proc %p (pid %d, %s)", p, p->p_pid,
776 if (PCPU_GET(switchtime.tv_sec) == 0)
777 microuptime(PCPU_PTR(switchtime));
778 PCPU_SET(switchticks, ticks);
779 mtx_unlock_spin(&sched_lock);
782 * cpu_set_fork_handler intercepts this function call to
783 * have this call a non-return function to stay in kernel mode.
784 * initproc has its own fork handler, but it does return.
786 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
790 * Check if a kernel thread misbehaved and returned from its main
794 if (p->p_flag & P_KTHREAD) {
797 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
798 p->p_comm, p->p_pid);
804 crfree(td->td_ucred);
808 mtx_assert(&Giant, MA_NOTOWNED);
812 * Simplified back end of syscall(), used when returning from fork()
813 * directly into user mode. Giant is not held on entry, and must not
814 * be held on return. This function is passed in to fork_exit() as the
815 * first parameter and is called when returning to a new userland process.
818 fork_return(td, frame)
820 struct trapframe *frame;
823 userret(td, frame, 0);
825 if (KTRPOINT(td->td_proc, KTR_SYSRET)) {
826 ktrsysret(td->td_proc->p_tracep, SYS_fork, 0, 0);
829 mtx_assert(&Giant, MA_NOTOWNED);